Document Type

Article

Date of Original Version

2017

Abstract

Great care is generally taken in epidemiologic studies to ensure the internal validity of causal effect estimates; however, the external validity of effect estimates, has received considerably less attention. The causal effect in a given target population is the average of heterogeneous subgroup effects, weighted according to the prevalence of the subgroups in the target population. When the study sample is not a random sample of the target population, the sample average treatment effect, even if internally valid, cannot be expected to equal the average treatment effect in the target population. There are several categories of choices for the target population. The study sample may be a census of the target population; the population from which the study sample is a random sample or from which the study sample is not a random sample; or some other population of which, the study sample is not a subset of the target population. The identification conditions sufficient for external validity closely parallel the identification conditions for internal validity, namely: conditional exchangeability; positivity; similar distributions of the versions of; similar patterns of interference; no measurement error; and correct model specification. The value of an effect estimate for planning purposes and decision making will depend on the degree of departure from both internal and external validity. If the study sample is not a random sample of the target population, direct standardization (the g-formula or transport formula) or inverse probability weighting can be used to estimate a causal effect in the target population.

Epidemiology as a discipline is distinguished by its efforts to identify causes of disease for the purpose of intervening to improve public health. Great care is generally taken in epidemiologic studies to ensure the internal validity of causal effect estimates,1 including the application of methods to minimize the potential for bias due to measurement error, confounding, selection (specifically, due to missing data, including censoring and truncation), and model misspecification. However, the external validity of effect estimates, has received considerably less attention. For the purposes of this discussion, we use the term external validity to refer to the potential for an internally valid treatment (or exposure or intervention) effect measured in a study sample to differ from the treatment effect that would have been estimated in the population of interest2 (henceforth, the target population). External validity encompasses generalizability and transportability, which we distinguish below. We advance the discussion of external validity herein using a potential outcomes framework. We enumerate a set of identification assumptions sufficient to estimate an externally valid effect, and note the parallel between these and the identification assumptions sufficient to estimate an internally valid effect. Finally, we illustrate some issues regarding generalizability with a simple example and discuss practical considerations for addressing generalizability in epidemiological study design.

Share

COinS